|Title||Numerical Analysis of NO and CO in a Flameless Burner|
|Contributor||George Mallouppas, Rajesh Rawat|
|Subject||numerical analysis, NO, CO, Flameless burner, CD-adapco, Germany, London, Melville, New York, AFRC, September 2016, Kauai|
AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 Numerical analysis of NO and CO in a flameless burner Carlo Locci,1∗ George Mallouppas,2 Rajesh Rawat3 1 2 3 CD-adapco, Nordostpark 3-5, Nuremberg, 904111, Germany CD-adapco, 200 Shepherds Bush Road, London, W6 7NL, UK CD-adapco, 60 Broadhollow Road, Melville, New York 11747, US ∗ e-mail: email@example.com Combustion modification techniques represent a practical solution for burner design and optimization in reducing pollutant emissions. Flameless combustion (FC) is widely used to reduce emissions in confined furnaces due to its high efficiency and low emissions. In this work, Large Eddy Simulation (LES) modeling was applied to a small-scale combustor under FC operating conditions. Statistical analysis is performed and the results are compared with the available experimental data. The Flamelet Generated Manifold (FGM) model, which allows a computationally efficient combustion description is used to characterize combustion and auto-ignition. The FGM calculation results and trends are in good agreement with the available data. CO predictions match experiment within the measurement errors. Thermal and prompt NO are included in the calculations and compared with the data. The agreement between the calculation and the experiment is fair. 1 Introduction Combustion of fossil fuels provides up to 80% of the energy produced worldwide, leading to considerably large emissions of CO2 and NOx (1). Motivated by a more stringent legislation the engineering and research community are moving towards cleaner and more efficient technologies. CO2 is considered as the primary cause of global warming whilst CO and NOx are harmful for the human health and the local environment (2). For this reason, modern combustion #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 technologies aim to reduce CO, CO2 and NOx emissions with more efficient and less polluting designs. Emissions abatement systems are commonly divided into three main techniques: i) pre-combustion, ii) post-combustion, and iii) combustion modification. In pre-combustion techniques, cleaner fuels are used to reduce the fuel bound NOx . Oxy-combustion also belongs to this group, where oxygen is separated from the air prior to combustion. On the other hand, in post-combustion techniques exhaust gases are treated with pollutant reducing agents or filters to meet environmental emissions restrictions. Common designs for combustion modification include low excess air techniques, air or fuel staging, over-fire air reburning and flameless combustion (3, 4). This work focuses on the third technique using flameless combustion (FC) (5, 6). FC is an attractive solution because it can be an efficient technology due to its very low CO, NOx and soot emissions. In FC burners, burnt gases are recirculated at a high rate, with preheated oxidizer to sustain combustion and also avoid quenching. The oxygen content is low, resulting in reduced and smoothed local temperature peaks. This leads to a twofold benefit: i) the general efficiency of the process is increased, and ii) the thermal NOx are reduced due to the lower temperatures. Computational Fluid Dynamics (CFD) is a valuable tool to improve such designs with the additional benefit of low development costs. Two main aspects are critical for high quality numerical prediction of FC burners, namely the slow auto-ignition process due to the dilution of burnt gases and the turbulent mixing of the fuel/air/burnt gases. In this paper, a flameless combustor burner is numerically examined and the results are scrutinized with the available experimental data. The experimental data are well characterized and cover a range of operating conditions. The objective of this work is to evaluate the tabulated chemistry Flamelet Generated Manifold (FGM) model in designing FC burners. Numerical modeling is performed with the commercial software STAR-CCM+ v.11.02 and chemistry is computed and tabulated with DARS v3.02. High fidelity calculations in the context of Large Eddy Simulations (LES) are performed to obtain a better representation of the turbulent mixing due to the small turbulent scales. The model is used to examine the formation of CO and NO emissions. The advantages/disadvantages of the combustion model will be discussed including the benefits of LES that is able to capture the important mixing scales critical in predicting the correct flame dynamics and emissions. #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 2 2.1 Numerical models Flow modelling In the current work, LES is used to perform the calculations. Briefly, in LES calculations a spatial filter is applied to the Navier-Stokes equations, leading to a fully resolved simulation of the larger turbulent eddies. Smaller eddies are accounted for through a sub-grid model function of the turbulent viscosity µt . This approach is more expensive compared to Reynolds Averaged Navier-Stokes (RANS), due to he spatial and temporal resolution required. and ∂ ρ̄ ∂ ρ̄ũi + =0 ∂t ∂xi (1) ∂τijsgs ∂ ρ̄ũj ∂ ρ̄ũj ũi ∂ P̄˜ ∂ τ̄˜ij + =− + − ∂t ∂xi ∂xj ∂xi ∂xi (2) where ρ̄ is the Favre averaged density, ũj is the density weighted filtered velocity, τijsgs is the subgrid scale shear stresses which are calculated with the WALE subgrid scale model. The WALE model uses a mixing-length type formula to model τijsgs (7). The turbulent viscosity µt is given as µt = ρ̄∆2 SW (3) where SW is the deformation parameter and ∆ is the length scales or grid filter width in terms of the cell volume V . ∆ is given as h ∆ = min CW V 1/ 3 i , κd (4) where κ is the von Karman constant (0.41) and d is the wall distance. SW is defined as: 3/ 2 SW = Sd : Sd 5/ 4 Sd : Sd + S : S 5 /2 (5) where S is the strain tensor and the tensor Sd is defined as: Sd = h iT #17; 1#16; 5 ũ • 5ũ + 5 ũ • 5ũ 2 (6) The coefficient CW in the current work is set to 0.544. Typical values of CW are 0.5 for homogeneous isotropic decaying turbulence to 0.325 for channel flows. Validations using STAR-CCM+ have shown that the WALE model seems seemingly less sensitive to CW . Additionally, 0.544 works very well for both homogeneous isotropic decaying turbulence and for channel flows. The advantage of the WALE model is that it automatically gives accurate scaling at the walls. #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 2.2 Combustion modeling In the present work the FGM model is scrutinised against the experimental data. This section briefly presents the FGM framework.In addition, the NO models will be briefly discussed. 2.2.1 Flamelet Generated Manifold The FGM model is a tabulated chemistry approach. Tabulated chemistry allows fast computations of the reactive flow in a calculation. In particular, tabulated chemistry models use a 1D simplified chemical system that are simulated as a function of a set of functional parameters. The FGM model uses 0D ignition manifold that are tabulated as a function of mixture fraction, enthalpy and progress variable to track auto-ignition. This aspect is fundamental in simulating FC, as this combustion mode is often described as a slow auto-ignition phenomena triggered by the burnt gases which dilute the fuel/air mixture (6). Following Bilger, the mixture fraction is defined as follows (8): Z= f −f f −f fC −fC,ox + 0.5 H mHH,ox − O mOO,ox mC f −f f −f f −f 2 C,fmCC,ox + 0.5 H,fmHH,ox − O,fmOO,ox 2 (7) where f is the element mass fraction, subscripts ox and f represent the oxidizer and fuel streams and m is atomic mass. It is generally normalized and has a unitary value at the fuel stream and is equal to 0 in the oxidizer stream. The FGM manifold is exploited in the LES simulation through the coupling of the transport equations mentioned above. Mixture fraction being a passive scalar, is only convected and diffused without any source term. Its transport equation before filtering due to LES is: ∂ρZ + ∇ · (ρuZ) − ∇ · (ΓZ ∇Z ) = 0 ∂t (8) where Γ is the diffusive term. Progress variable equation is: ∂ρYc + ∇ · (ρuYc ) − ∇ · (ΓYc ∇Yc ) = ω̇Yc ∂t (9) and includes the source term ω̇Yc which is directly calculated from the FGM manifold. Auto-ignition is tracked with a progress variable defined in this work as the sum of CO and CO2 mass fractions. Enthalpy is artificially increased and decreased in the reactor to include over and under adiabatic conditions respectively. Turbulence-chemistry interaction can be accounted for through the mixture fraction variance Zv , which quantifies the local turbulent fluctuation in the #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 (a) (b) (c) Figure 1: Temperature from the FGM manifold for the simulations of this work as a function of the mixture fraction variance Zv = 0 and adiabatic conditions. Plot as a function of mixture fraction Z and normalized progress variable c. Z space. Notice that progress variable fluctuations can also be accounted for (9), although were neglected for this study as done in (10). An example of the FGM manifold for the conditions of this work is shown in Figs. 1 and 2, where temperature is plotted as a function of mixture fraction Z and the normalized progress variable c. All the manifolds of this work are calculated with the GRI 3.0 mechanism (11). The impact of Zv is observed in Fig. 1, with a significant smoothing of temperature gradients as the variance increases. For Zv = 0, the highest temperature is reached at the equilibrium value of c = 1 and around the stoichiometric Z. As Zv increases, stoichiometric temperature decreases due to the lean and rich regions contribution, thus imitating the local fluctuations of mixture fraction. In the same way, rich and lean regions temperature increases due to the stoichiometric temperature influence. Heat loss is artificially obtained by diminishing the enthalpy in the manifold and accounts for the possibility of walls cooling or radiation. As shown in Fig. 2, heat loss leads to a generic decrease of temperature with no significant smoothing of the gradients. A significant heat loss can trigger the quenching of the manifold. 2.2.2 NO emissions modeling Thermal NO is also simulated with a tabulated model for FGM. A transport equation is solved for NO, with the production rate calculated as follows (12): RN O = A − B[N O] + C AD [N O] + C D − [N O]2 B [N O] + C D (10) #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 (a) (b) (c) Figure 2: Temperature from the FGM manifold for the simulations of this work as a function of the heat loss level and Zv = 0. Plot as a function of mixture fraction Z and normalized progress variable c. where [N O] is the molar concentration of NO and A, B, C and D are tabulated coefficients function of O, H and OH radicals molar concentration and thermal NO reaction rate from Zeldovich mechanism (13). Prompt NO is calculated as a function of the Fuel (in this case CH4 ) and O2 species, depending on the formulation proposed in (14). Note that CO is directly evaluated from the tables. 3 Configuration set-up The burner of Verissimo et al. (15) is chosen to evaluate FGM in the context of FC mode. The 10 kW small-scale laboratory burner is presented in Fig. 3 and uses methane as fuel. Methane is fed in from 16 fuel injectors of 2 mm of diameter, which surround a main air jet. Combustion is stabilized by heating air at 673 K whilst burnt gas recirculation is assured with an internal recirculation movement, generated by the high momentum of the central jet. To ensure complete combustion of methane, the burner operates at lean conditions with a fixed air to fuel ratio of 1.3. The burner is mainly cylindrical with a diameter of 100 mm and an axial length of 340 mm. The flow is then accelerated at the end of the burner via a 150 mm length convergent nozzle. Note that radiation was not included in the current calculations because heat losses from the wall are the major cause for the enthalpy loss and correspond to approximately half of the energy input (10). #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 Figure 3: Schematic representation of the Verissimo burner. 3.1 Experimental measurement stations Centerline and radial experimental measurements are available for temperature, and major and minor species at the positions highlighted in Fig. 4. In this work we focused on the prediction of CO, CO2 and NO. Notice that velocity was not measured for the flameless configuration. 3.2 Mesh details Polyhedral volume elements were used to build the mesh. As it can be observed in Fig. 4, the mesh is refined in the near fuel nozzle region to better describe methane entrainment in the air jet. This refinement is based on velocity flow field which was determined from a RANS steady state calculation. The second refined region was individuated as a function of the mixture fraction profiles and includes all the zones where Z gradients are still large. The minimum cell size of the refined region was set to 0.5 mm whilst external regions were less refined with elements on the order of 3 mm. Gradients stabilization was ensured with smooth cell size increase. Note that 2 prism wall layers were added. In total, ∼6.3 M polyhedral volume elements were used. 3.3 Boundary and initial conditions The LES calculations are initialised from the RANS steady-state solution in order to speed-up the calculations. Table 1 summarises the boundary conditions used in this work. As already mentioned, the FC burner operates at very lean conditions (1.3 air to fuel ratio). Air is preheated at 673.15 K and fuel is injected at room temperature. Finally, the FC burner operates at atmospheric conditions. #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 Figure 4: Geometric details of the Verissimo burner and the grid refinement strategy used for the simulations. The dashed blue line correspond to the centreline and radial measurements locations. Table 1: Table summarizes the boundary conditions. Quantity Mass flow rate (kg/s) Temperature (K) Species mass fractions (-) Turbulent intensity (%) Turbulent length scale (mm) 3.4 Air Inlet 4.59×10−3 673.15 O2 = 0.233, N2 = 0.767 10.0 1.429 Fuel Inlet 2.06×10−4 298.15 CH4 = 1.0 Wall in combustion chamber 1100.0 - 10.0 0.286 - Choice of the wall temperature As wall temperature Tw was not measured, preliminary tests were carried out to assess its value on the walls depending on NO prediction. Sensitivity of NO to temperature was addressed in Fig. 5 in the RANS context, where the axial profiles of temperature and NO are shown as a function of Tw . It is here underlined the utility of RANS techniques for such studies, as global information on the burner can be retrieved at a lower cost. In Fig. 5 the temperature profiles vary little as a function of Tw and are less accurate if compared to LES results presented in the next sections, due to the less accurate mixing prediction of RANS. The final equilibrium values are however correctly predicted, as they depend less on the mixing but rather on the thermodynamic balance of the furnace. If temperature varies little depending on wall temperature, NO profiles are extremely sensitive to this parameter. Following these tests, we set Tw to 1100 K, which is consistent to other previous works (10), (16). #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 Figure 5: RANS results of the sensitivity analysis for NO as a function of the wall temperature Tw . 3.5 Results and flame structure analysis The FGM results are compared to the axial experimental profiles in Fig. 6 for temperature, major species CO2 , CO and NO. As expected LES is able to capture both the temperature increase as opposed to RANS. It can be noticed that temperature profiles are typical of FC, with no visible peak after a slow temperature increase due to dilution. CO2 results capture the tendency and the equilibrium final experimental values. Compared to temperature, the agreement is less satisfactory in the first part of the burner. This point was already studied in previously works. Locci et al. (10) found that no equilibrium compositions can be equivalent to the experimental conditions in the first part of the burner. Cuoci et al. (17) and Lamouroux et al. (16) suspected a problem in the measurements, because of the use of invasive probes in the high momentum region. CO peak prediction is shifted more upstream than the experimental results. Such result can be tied to a poor mixing prediction which leads to an earlier auto-ignition of the fresh gases. However, the results of temperature and CO2 correctly capture the flame increase. The second hypothesis is related to the FGM formulation which presents some limitation in simulating FC. In this combustion mode, dilution plays an important role and was accounted for in previous work related to tabulated chemistry either in the PSR (10) or diffusion flamelet (16) contexts. It should be noticed however, that including dilution in the formulation, leads to a more complex model with a higher computational costs. Finally, turbulence intensity at the air inlet might play a role in correctly capturing CO peak. Contrarily to CO2 and temperature, CO is the only variable to present a clear peak. This behavior is tied to the extreme sensitivity of CO to enthalpy loss, which still occurs in the downstream positions because of the flow cooling from the walls. NO results are satisfactory with a 1.5 factor with respect to experimental data. This result can be compared to the tabulated approach of Locci et al. (18), where #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 Figure 6: Comparison of centreline results of Temperature, CO2 , CO and NO with the available experimental data. similar results where obtained in the diffusion flamelet context. 4 ppm of NO where measured axially for x=11 mm. In (18) it was argued that this could be due to experimental uncertainty, as CO2 and CO are equal to 0 at the same position, thus excluding the possibility of burnt gases dilution in that region. LES contour field of velocity and temperature are presented in Fig.7. Temperature field confirm the typical trend of FC, with a very slow transition from fresh gases to burnt gases. The flame is not defined and burnt gases diffuse inside the chamber. The temperature delta in FC burners is small, with an increase of approximately 1000 K compared to a diffusion flame, where temperature increase can be much larger (19). Due to the high-momentum jet, the central flow spreads in the chamber and impacts the walls. This creates lateral recirculation regions which convect burnt gases back to the fresh gases region. The iso-line of V = 0[m/s] in Fig.7, shows the extent of the recirculation region which convect a part of the burnt gases back to the fresh gases region. The radial profiles for the measured variables are presented in Figs. 8, 9 and 10. On overall, radial profiles of temperature and CO2 clearly show a FC mode, with very flat profiles around the flame. Both CO2 and temperature radial profiles are correctly predicted although temperature is slightly overestimated. This point was already addressed in previous work and since wall temperature was not established experimentally, it is hard to understand how to improve the quality of temperature prediction. The near jet predictions of CO are overestimated of two #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 Figure 7: Temperature (top) and velocity (bottom) FGM-LES contour field of the simulated burner. Iso-lines of velocity=0 m/s are also shown to individuate recirculation regions. Figure 8: Radial plots of FGM-LES results of temperature, CO2 , CO and NO at x = 11 mm and x = 45 mm from the nozzle. #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 Figure 9: Radial plots of FGM-LES results of temperature, CO2 , CO and NO at x = 79 mm and x = 113 mm from the nozzle Figure 10: Radial plots of FGM-LES results of temperature, CO2 , CO and NO at x = 147 mm and x = 310 mm from the nozzle #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 Figure 11: Comparison of peak heat release with the available experimental data and the FGM model. orders of magnitude. Previous works already dealt with CO prediction with PSR manifolds. In particular Fiorina et al. (20) showed a tendency of such manifolds to overpredict CO for rich mixtures in the non-premixed context. Although FGM models proved to be accurate enough in the non-premixed and partially premixed context (21), for this particular case, such large overestimation can be attributed to the absence of dilution as an additional parameter in the model. As already observed in the axial results of Fig. 6, CO peak is predicted more upstream. The issue in predicting the flame position is confirmed by OH fields, which are presented in Fig. 11 compared to the experimental results. OH is a radical intermediate species which is found across the flame front. It can be observed that OH is placed more downstream than experiments, probably due to the absence of dilution in formulation of this work. The final equilibrium values of the flame are correctly predicted, also for CO. Although FGM formulation is not suitable to predict CO in strongly diluted front flames, the equilibrium values are correctly predicted, proving that this model can be still useful to retrieve information in the design of industrial furnaces with a very contained computational cost compared to more complex formulations. #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 4 Conclusion The tabulated chemistry FGM formulation was applied to a small-scale flameless burner. The flameless burner presented a simple geometry with a internal recirculation pattern to favor dilution from burnt gases. Experimental data were available for temperature, CO2 , CO and NO. Mixing and auto-ignition both play a major role in FC. Mixing prediction was accounted for by simulating the furnace with a LES formulation, which is more accurate compared to RANS. Auto-ignition was accounted for in the FGM model with a progress variable to track temperature and combustion products in PFR reactors. Both thermal and prompt NO were calculated with a tabulated approach, based on complex chemistry. Experimental data presented some uncertainties although can be considered reliable enough for such study. A part for CO, correct results for all the variables were obtained with a satisfactory agreement for both axial and radial profiles. FGM formulation showed some limitation in predicting CO for strongly diluted combustion, with a large overestimation of this species in rich regions and a shifted peak. The final experimental value was however correctly predicted, thus concluding that the model is suitable for burner design, when targeting CO optimization. Finally, also NO were successfully predicted with results comparable to previous studies. References and Notes 1. Internation Energy Agency. Key world energy statistics. 2012. 2. J. Liang. Chemical modeling for air resources fundamentals, applications, and corroborative analysis. Elsevier.Inc, ISBN: 978-0-12-408135-2, 2013. 3. Hill S, Smoot L. Modeling of nitrogen oxides formation and destruction in combustion systems. Prog Energy Combust Sci 2000;26:417-458. 4. Skalska K, Miller JS, Ledakowicz S. Trends in NOx abatement: A review. Sci Total Environ 2010;408:3976-3989. 5. J.A. Wunning; J.G. Wunning. Flameless oxidation to reduce thermal no-formation. Prog. Energy Combust. Sci., 23:8194, 1997. 6. A. Cavaliere; M. de Joannon. Mild combustion. Prog. Energy Combust. Sci., 30:329366, 2004. 7. F. Nicoud; F. Durcos. Subgrid-Scale Stress Modelling Based on the Square of the Velocity Gradient Tensor. Flow, Turbulence and Combustion, 62, pp. 183-200. 1999. #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 8. R. W. Bilger. Proc. Combust. Inst. 47-488. 1988. 9. L. Vervisch; R. Hauguel; P. Domingo; R. Rullaud. Three facets of turbulent combustion modeling: Dns of premixed v-ame, les of lifted nonpremixed ame and rans of jetame. Journ. Turb., 5, 2004. 10. C. Locci; O. Colin; J.B. Michel. Large eddy simulations of a small-scale ameless combustor by means of diluted homogeneous reactors. 93 2014. 11. G.P. Smith; D.M. Golden; M. Frenklach; N.W. Moriarty; B. Eiteneer; M. Goldenberg; C.T. Bowman; R.K. Hanson; S. Song; W.C. Jr. Gardiner; V.V. Lissianski; Z. Qin. http : //www.me.berkely.eud/grimech/. 12. Baulch, D. L., Drysdall, D. D., Horne, D. G., and Lloyd, A. C. 1973. Evaluated Kinetic Data for High Temperature Reactions, vols. 1-3. Butterworth. 13. Y.B. Zeldovich. Acta. Physicochim., 21:577628, 1946. 14. De Soete, G. G. 1975. Overall Reaction Rates of NO and N2 Formation from Fuel Nitrogen, in 15th Symp. (Intl.) on Combustion, pp. 1093-1102. The Combustion Institute. 15. A.S. Verissimo; A.M.A Rocha; M. Costa. Operational, combustion, and emission characteristics of a small-scale combustor. Ener. Fuel., 25:24692480, 2011. 16. J.Lamouroux; M.Ihme; B.Fiorina; O.Gicquel. Tabulated chemistry approach for diluted combustion regimes with internal recirculation and heat losses. Combust. Flame, 161, 2014. 17. A. Cuoci; A. Frassoldati; A. Stagni; T. Faravelli; E. Ranzi; G. Buzzi-Ferraris. Numerical modeling of nox formation in turbulent ames using a kinetic postprocessing technique. Ener. Fuel., 27:11041122, 2013. 18. C. Locci; O. Colin; D. Poitou; F. Mauss. A Tabulated, Flamelet Based No Model for Large Eddy Simulations of Non Premixed Turbulent Jets with Enthalpy Loss. Flow Turb. Comb. 94, 2015. 19. T. Poinsot; D. Veynante. Theoretical and numerical combustion. 3rd edition. 2012. 20. B.Fiorina; O. Gicquel; L. Vervisch; S.Carpentier; N. Darabiha. Approximating the chemical structure of partially premixed and diffusion counterow ames using fpi amelet tabulation. Combust. Flame., 140:147160, 2005. #12;AFRC 2016 Industrial Combustion Symposium Kauai, Hawaii, USA, September 11th - 14th , 2016 21. A.W. Vreman; B.A. Albrecht; J. van Oijen; L.P.H. de Goey; R.J.M Bastiaans. Premixed and nonpremixed generated manifolds in large-eddy simulation of sandia ame d and f. Combust. Flame., 153:394416, 2008.
|Metadata Cataloger||Catrina Wilson|